I had a feeling that I would get more stale shares using more than one miner per GPU. Also one should only trust the hashrate measured at the pool(s).
I run 2 Phoenix clients per GPU (I have 3 x HD5870) and I run them both at the same Aggression. This splits the work between them evenly until one of the pools has a hiccup. My total hash rate is slightly higher than running one instance of Phoenix per GPU.
Why would you get more stale shares using more than one miner per GPU?
The hash rate at the pool is an estimate, so nowhere near as accurate as the client reported hash rate.
I did the same thing happily for a while but now I am not so sure any more if it is the optimum.
Theoretically you get more stales because your workers are slower compared to the other workers in the pool. I get lots of stales but there might be some other reason I have yet to find.
---edit---
For my system the gain in hashrate using two miners is three to five times larger than the loss from stales. I will keep running two miners per GPU.
This is what I do. The only thing I've noticed is that I've had some disconnection problems arise when running multiple miner connections when I have had two computers both running two instances (four total connections) through the same mobile broadband connection. I don't know if it is a port issue with the router or what exactly, but it is something to be aware of.